hw/net/vhost_net.c

   1 /*
   2  * vhost-net support
   3  *
   4  * Copyright Red Hat, Inc. 2010
   5  *
   6  * Authors:
   7  *  Michael S. Tsirkin <mst@redhat.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  * Contributions after 2012-01-13 are licensed under the terms of the
  13  * GNU GPL, version 2 or (at your option) any later version.
  14  */
  15
  16 #include "qemu/osdep.h"
  17 #include "net/net.h"
  18 #include "net/tap.h"
  19 #include "net/vhost-user.h"
  20
  21 #include "hw/virtio/virtio-net.h"
  22 #include "net/vhost_net.h"
  23 #include "qemu/error-report.h"
  24
  25
  26 #ifdef CONFIG_VHOST_NET
  27 #include <linux/vhost.h>
  28 #include <sys/socket.h>
  29 #include <linux/kvm.h>
  30 #include <netpacket/packet.h>
  31 #include <net/ethernet.h>
  32 #include <net/if.h>
  33 #include <netinet/in.h>
  34
  35
  36 #include "standard-headers/linux/virtio_ring.h"
  37 #include "hw/virtio/vhost.h"
  38 #include "hw/virtio/virtio-bus.h"
  39
  40 struct vhost_net {
  41     struct vhost_dev dev;
  42     struct vhost_virtqueue vqs[2];
  43     int backend;
  44     NetClientState *nc;
  45 };
  46
  47 /* Features supported by host kernel. */
  48 static const int kernel_feature_bits[] = {
  49     VIRTIO_F_NOTIFY_ON_EMPTY,
  50     VIRTIO_RING_F_INDIRECT_DESC,
  51     VIRTIO_RING_F_EVENT_IDX,
  52     VIRTIO_NET_F_MRG_RXBUF,
  53     VIRTIO_F_VERSION_1,
  54     VIRTIO_NET_F_MTU,
  55     VIRTIO_F_IOMMU_PLATFORM,
  56     VHOST_INVALID_FEATURE_BIT
  57 };
  58
  59 /* Features supported by others. */
  60 static const int user_feature_bits[] = {
  61     VIRTIO_F_NOTIFY_ON_EMPTY,
  62     VIRTIO_RING_F_INDIRECT_DESC,
  63     VIRTIO_RING_F_EVENT_IDX,
  64
  65     VIRTIO_F_ANY_LAYOUT,
  66     VIRTIO_F_VERSION_1,
  67     VIRTIO_NET_F_CSUM,
  68     VIRTIO_NET_F_GUEST_CSUM,
  69     VIRTIO_NET_F_GSO,
  70     VIRTIO_NET_F_GUEST_TSO4,
  71     VIRTIO_NET_F_GUEST_TSO6,
  72     VIRTIO_NET_F_GUEST_ECN,
  73     VIRTIO_NET_F_GUEST_UFO,
  74     VIRTIO_NET_F_HOST_TSO4,
  75     VIRTIO_NET_F_HOST_TSO6,
  76     VIRTIO_NET_F_HOST_ECN,
  77     VIRTIO_NET_F_HOST_UFO,
  78     VIRTIO_NET_F_MRG_RXBUF,
  79     VIRTIO_NET_F_MTU,
  80     VIRTIO_F_IOMMU_PLATFORM,
  81
  82     /* This bit implies RARP isn't sent by QEMU out of band */
  83     VIRTIO_NET_F_GUEST_ANNOUNCE,
  84
  85     VIRTIO_NET_F_MQ,
  86
  87     VHOST_INVALID_FEATURE_BIT
  88 };
  89
  90 static const int *vhost_net_get_feature_bits(struct vhost_net *net)
  91 {
  92     const int *feature_bits = 0;
  93
  94     switch (net->nc->info->type) {
  95     case NET_CLIENT_DRIVER_TAP:
  96         feature_bits = kernel_feature_bits;
  97         break;
  98     case NET_CLIENT_DRIVER_VHOST_USER:
  99         feature_bits = user_feature_bits;
 100         break;
 101     default:
 102         error_report("Feature bits not defined for this type: %d",
 103                 net->nc->info->type);
 104         break;
 105     }
 106
 107     return feature_bits;
 108 }
 109
 110 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
 111 {
 112     return vhost_get_features(&net->dev, vhost_net_get_feature_bits(net),
 113             features);
 114 }
 115
 116 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
 117 {
 118     net->dev.acked_features = net->dev.backend_features;
 119     vhost_ack_features(&net->dev, vhost_net_get_feature_bits(net), features);
 120 }
 121
 122 uint64_t vhost_net_get_max_queues(VHostNetState *net)
 123 {
 124     return net->dev.max_queues;
 125 }
 126
 127 uint64_t vhost_net_get_acked_features(VHostNetState *net)
 128 {
 129     return net->dev.acked_features;
 130 }
 131
 132 static int vhost_net_get_fd(NetClientState *backend)
 133 {
 134     switch (backend->info->type) {
 135     case NET_CLIENT_DRIVER_TAP:
 136         return tap_get_fd(backend);
 137     default:
 138         fprintf(stderr, "vhost-net requires tap backend\n");
 139         return -EBADFD;
 140     }
 141 }
 142
 143 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 144 {
 145     int r;
 146     bool backend_kernel = options->backend_type == VHOST_BACKEND_TYPE_KERNEL;
 147     struct vhost_net *net = g_new0(struct vhost_net, 1);
 148     uint64_t features = 0;
 149
 150     if (!options->net_backend) {
 151         fprintf(stderr, "vhost-net requires net backend to be setup\n");
 152         goto fail;
 153     }
 154     net->nc = options->net_backend;
 155
 156     net->dev.max_queues = 1;
 157     net->dev.nvqs = 2;
 158     net->dev.vqs = net->vqs;
 159
 160     if (backend_kernel) {
 161         r = vhost_net_get_fd(options->net_backend);
 162         if (r < 0) {
 163             goto fail;
 164         }
 165         net->dev.backend_features = qemu_has_vnet_hdr(options->net_backend)
 166             ? 0 : (1ULL << VHOST_NET_F_VIRTIO_NET_HDR);
 167         net->backend = r;
 168         net->dev.protocol_features = 0;
 169     } else {
 170         net->dev.backend_features = 0;
 171         net->dev.protocol_features = 0;
 172         net->backend = -1;
 173
 174         /* vhost-user needs vq_index to initiate a specific queue pair */
 175         net->dev.vq_index = net->nc->queue_index * net->dev.nvqs;
 176     }
 177
 178     r = vhost_dev_init(&net->dev, options->opaque,
 179                        options->backend_type, options->busyloop_timeout);
 180     if (r < 0) {
 181         goto fail;
 182     }
 183     if (backend_kernel) {
 184         if (!qemu_has_vnet_hdr_len(options->net_backend,
 185                                sizeof(struct virtio_net_hdr_mrg_rxbuf))) {
 186             net->dev.features &= ~(1ULL << VIRTIO_NET_F_MRG_RXBUF);
 187         }
 188         if (~net->dev.features & net->dev.backend_features) {
 189             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 190                    " for backend\n",
 191                    (uint64_t)(~net->dev.features & net->dev.backend_features));
 192             goto fail;
 193         }
 194     }
 195
 196     /* Set sane init value. Override when guest acks. */
 197     if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
 198         features = vhost_user_get_acked_features(net->nc);
 199         if (~net->dev.features & features) {
 200             fprintf(stderr, "vhost lacks feature mask %" PRIu64
 201                     " for backend\n",
 202                     (uint64_t)(~net->dev.features & features));
 203             goto fail;
 204         }
 205     }
 206
 207     vhost_net_ack_features(net, features);
 208
 209     return net;
 210
 211 fail:
 212     vhost_dev_cleanup(&net->dev);
 213     g_free(net);
 214     return NULL;
 215 }
 216
 217 static void vhost_net_set_vq_index(struct vhost_net *net, int vq_index)
 218 {
 219     net->dev.vq_index = vq_index;
 220 }
 221
 222 static int vhost_net_start_one(struct vhost_net *net,
 223                                VirtIODevice *dev)
 224 {
 225     struct vhost_vring_file file = { };
 226     int r;
 227
 228     net->dev.nvqs = 2;
 229     net->dev.vqs = net->vqs;
 230
 231     r = vhost_dev_enable_notifiers(&net->dev, dev);
 232     if (r < 0) {
 233         goto fail_notifiers;
 234     }
 235
 236     r = vhost_dev_start(&net->dev, dev);
 237     if (r < 0) {
 238         goto fail_start;
 239     }
 240
 241     if (net->nc->info->poll) {
 242         net->nc->info->poll(net->nc, false);
 243     }
 244
 245     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 246         qemu_set_fd_handler(net->backend, NULL, NULL, NULL);
 247         file.fd = net->backend;
 248         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 249             r = vhost_net_set_backend(&net->dev, &file);
 250             if (r < 0) {
 251                 r = -errno;
 252                 goto fail;
 253             }
 254         }
 255     }
 256     return 0;
 257 fail:
 258     file.fd = -1;
 259     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 260         while (file.index-- > 0) {
 261             int r = vhost_net_set_backend(&net->dev, &file);
 262             assert(r >= 0);
 263         }
 264     }
 265     if (net->nc->info->poll) {
 266         net->nc->info->poll(net->nc, true);
 267     }
 268     vhost_dev_stop(&net->dev, dev);
 269 fail_start:
 270     vhost_dev_disable_notifiers(&net->dev, dev);
 271 fail_notifiers:
 272     return r;
 273 }
 274
 275 static void vhost_net_stop_one(struct vhost_net *net,
 276                                VirtIODevice *dev)
 277 {
 278     struct vhost_vring_file file = { .fd = -1 };
 279
 280     if (net->nc->info->type == NET_CLIENT_DRIVER_TAP) {
 281         for (file.index = 0; file.index < net->dev.nvqs; ++file.index) {
 282             int r = vhost_net_set_backend(&net->dev, &file);
 283             assert(r >= 0);
 284         }
 285     }
 286     if (net->nc->info->poll) {
 287         net->nc->info->poll(net->nc, true);
 288     }
 289     vhost_dev_stop(&net->dev, dev);
 290     vhost_dev_disable_notifiers(&net->dev, dev);
 291 }
 292
 293 int vhost_net_start(VirtIODevice *dev, NetClientState *ncs,
 294                     int total_queues)
 295 {
 296     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 297     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 298     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 299     int r, e, i;
 300
 301     if (!k->set_guest_notifiers) {
 302         error_report("binding does not support guest notifiers");
 303         return -ENOSYS;
 304     }
 305
 306     for (i = 0; i < total_queues; i++) {
 307         struct vhost_net *net;
 308
 309         net = get_vhost_net(ncs[i].peer);
 310         vhost_net_set_vq_index(net, i * 2);
 311
 312         /* Suppress the masking guest notifiers on vhost user
 313          * because vhost user doesn't interrupt masking/unmasking
 314          * properly.
 315          */
 316         if (net->nc->info->type == NET_CLIENT_DRIVER_VHOST_USER) {
 317             dev->use_guest_notifier_mask = false;
 318         }
 319      }
 320
 321     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, true);
 322     if (r < 0) {
 323         error_report("Error binding guest notifier: %d", -r);
 324         goto err;
 325     }
 326
 327     for (i = 0; i < total_queues; i++) {
 328         r = vhost_net_start_one(get_vhost_net(ncs[i].peer), dev);
 329
 330         if (r < 0) {
 331             goto err_start;
 332         }
 333
 334         if (ncs[i].peer->vring_enable) {
 335             /* restore vring enable state */
 336             r = vhost_set_vring_enable(ncs[i].peer, ncs[i].peer->vring_enable);
 337
 338             if (r < 0) {
 339                 goto err_start;
 340             }
 341         }
 342     }
 343
 344     return 0;
 345
 346 err_start:
 347     while (--i >= 0) {
 348         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 349     }
 350     e = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 351     if (e < 0) {
 352         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", e);
 353         fflush(stderr);
 354     }
 355 err:
 356     return r;
 357 }
 358
 359 void vhost_net_stop(VirtIODevice *dev, NetClientState *ncs,
 360                     int total_queues)
 361 {
 362     BusState *qbus = BUS(qdev_get_parent_bus(DEVICE(dev)));
 363     VirtioBusState *vbus = VIRTIO_BUS(qbus);
 364     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(vbus);
 365     int i, r;
 366
 367     for (i = 0; i < total_queues; i++) {
 368         vhost_net_stop_one(get_vhost_net(ncs[i].peer), dev);
 369     }
 370
 371     r = k->set_guest_notifiers(qbus->parent, total_queues * 2, false);
 372     if (r < 0) {
 373         fprintf(stderr, "vhost guest notifier cleanup failed: %d\n", r);
 374         fflush(stderr);
 375     }
 376     assert(r >= 0);
 377 }
 378
 379 void vhost_net_cleanup(struct vhost_net *net)
 380 {
 381     vhost_dev_cleanup(&net->dev);
 382 }
 383
 384 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 385 {
 386     const VhostOps *vhost_ops = net->dev.vhost_ops;
 387
 388     assert(vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
 389     assert(vhost_ops->vhost_migration_done);
 390
 391     return vhost_ops->vhost_migration_done(&net->dev, mac_addr);
 392 }
 393
 394 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 395 {
 396     return vhost_virtqueue_pending(&net->dev, idx);
 397 }
 398
 399 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 400                               int idx, bool mask)
 401 {
 402     vhost_virtqueue_mask(&net->dev, dev, idx, mask);
 403 }
 404
 405 VHostNetState *get_vhost_net(NetClientState *nc)
 406 {
 407     VHostNetState *vhost_net = 0;
 408
 409     if (!nc) {
 410         return 0;
 411     }
 412
 413     switch (nc->info->type) {
 414     case NET_CLIENT_DRIVER_TAP:
 415         vhost_net = tap_get_vhost_net(nc);
 416         break;
 417     case NET_CLIENT_DRIVER_VHOST_USER:
 418         vhost_net = vhost_user_get_vhost_net(nc);
 419         assert(vhost_net);
 420         break;
 421     default:
 422         break;
 423     }
 424
 425     return vhost_net;
 426 }
 427
 428 int vhost_set_vring_enable(NetClientState *nc, int enable)
 429 {
 430     VHostNetState *net = get_vhost_net(nc);
 431     const VhostOps *vhost_ops = net->dev.vhost_ops;
 432
 433     nc->vring_enable = enable;
 434
 435     if (vhost_ops && vhost_ops->vhost_set_vring_enable) {
 436         return vhost_ops->vhost_set_vring_enable(&net->dev, enable);
 437     }
 438
 439     return 0;
 440 }
 441
 442 int vhost_net_set_mtu(struct vhost_net *net, uint16_t mtu)
 443 {
 444     const VhostOps *vhost_ops = net->dev.vhost_ops;
 445
 446     if (!vhost_ops->vhost_net_set_mtu) {
 447         return 0;
 448     }
 449
 450     return vhost_ops->vhost_net_set_mtu(&net->dev, mtu);
 451 }
 452
 453 #else
 454 uint64_t vhost_net_get_max_queues(VHostNetState *net)
 455 {
 456     return 1;
 457 }
 458
 459 struct vhost_net *vhost_net_init(VhostNetOptions *options)
 460 {
 461     error_report("vhost-net support is not compiled in");
 462     return NULL;
 463 }
 464
 465 int vhost_net_start(VirtIODevice *dev,
 466                     NetClientState *ncs,
 467                     int total_queues)
 468 {
 469     return -ENOSYS;
 470 }
 471 void vhost_net_stop(VirtIODevice *dev,
 472                     NetClientState *ncs,
 473                     int total_queues)
 474 {
 475 }
 476
 477 void vhost_net_cleanup(struct vhost_net *net)
 478 {
 479 }
 480
 481 uint64_t vhost_net_get_features(struct vhost_net *net, uint64_t features)
 482 {
 483     return features;
 484 }
 485
 486 void vhost_net_ack_features(struct vhost_net *net, uint64_t features)
 487 {
 488 }
 489
 490 uint64_t vhost_net_get_acked_features(VHostNetState *net)
 491 {
 492     return 0;
 493 }
 494
 495 bool vhost_net_virtqueue_pending(VHostNetState *net, int idx)
 496 {
 497     return false;
 498 }
 499
 500 void vhost_net_virtqueue_mask(VHostNetState *net, VirtIODevice *dev,
 501                               int idx, bool mask)
 502 {
 503 }
 504
 505 int vhost_net_notify_migration_done(struct vhost_net *net, char* mac_addr)
 506 {
 507     return -1;
 508 }
 509
 510 VHostNetState *get_vhost_net(NetClientState *nc)
 511 {
 512     return 0;
 513 }
 514
 515 int vhost_set_vring_enable(NetClientState *nc, int enable)
 516 {
 517     return 0;
 518 }
 519
 520 int vhost_net_set_mtu(struct vhost_net *net, uint16_t mtu)
 521 {
 522     return 0;
 523 }
 524 #endif